Sample Size and Power Analysis
Workshop 18 - Winter Semester 2025-26
Overview
This workshop covers the principles of statistical power and sample size calculation for robust research design, with a focus on Bayesian sequential designs using HDI/ROPE criteria. The workshop explores practical approaches to determining sample sizes when both detection (PRESENT) and exclusion (ABSENT) decisions are required simultaneously.
Learning Objectives
- Understand statistical power fundamentals
- Calculate appropriate sample sizes
- Estimate and interpret effect sizes
- Apply power analysis tools in R
Topics Covered
- Statistical power fundamentals (Type I/II errors, power curves)
- Frequentist vs. Bayesian approaches to sample size determination
- Bayesian sequential designs with ROPE (Region of Practical Equivalence)
- HDI/ROPE decision rules for PRESENT and ABSENT verdicts
- Simulation-based power analysis
- Effect size estimation and sensitivity analysis
- Tools for power analysis in R (brms, bayestestR)
Materials
This workshop showcases an exploratory simulation study that demonstrates what's possible with Bayesian sequential designs. The example is based on ongoing research and serves as a practical illustration of the methodology rather than a finished tutorial.
Main Notebook
- Sequential Design Simulation Report - Interactive report showing a Bayesian sequential design for an iconic-duration effect study (4 scenarios: Conservative, Realistic, Optimistic, Minimal)
- GitHub Repository - Full simulation code and documentation
Background Material (Bayesian Workshops)
The sequential design relies heavily on concepts covered in the Bayesian modeling workshop series:
- ROPE (Region of Practical Equivalence) - Understanding ROPE-based decision criteria
- Bayesian Workshop Materials - Full workshop series on Bayesian regression with brms
Key Concepts
- Sequential design: Check for decisions at multiple checkpoints (N = 30, 45, 60, ..., 120) and stop early when evidence is decisive
- HDI/ROPE: Declare an effect PRESENT when the 95% HDI lies entirely above ROPE, ABSENT when entirely inside ROPE
- Dual decisions: Simultaneously requiring PRESENT (early window) and ABSENT (late window) decisions creates asymmetric power requirements that conventional power analysis cannot address
Prerequisites
- Basic understanding of statistics and hypothesis testing
- Familiarity with Bayesian concepts (priors, posteriors, HDI) helpful but not required
- Basic R knowledge recommended
Instructor
Job Schepens
Project S, SFB 1252
University of Cologne
Session Details
Date: 4 March 2026
Time: 14:00 - 15:30
Location: House of Prominence, Attic, Luxemburger Str. 299, Cologne